Depth Map Acquisition From Camera Array

People use their eyes to perceive the three-dimensional world and obtain the most intuitive three-dimensional information.The various scenes in the life recorded by the camera are two-dimensional and cannot capture the three-dimensional information of the object through the photo.Stereo vision usually includes binocular stereo vision and multi-view stereo vision.Specific methods for obtaining depth using binocular or multi-cameras include active methods and passive methods.The passive methods relying on the camera itself are the focus of this research.Although it has the disadvantage of relatively sparse depth information,it is more suitable for rough depth information acquisition of outdoor scenes.It has already begun to be used in today's hot areas such as driverless,robot vision,robot obstacle avoidance,and virtual reality.This dissertation uses an array camera with three visible light cameras to collect scene information.It mainly focus on the fusion of depth image and the acquisition of disparity image.The research includes multi-cameras' calibration,stereo correction,stereo matching,depth image fusion,and so on.Through the use of multi array camera lens to obtain more accurate depth image,it can effectively deal with the depth image of the occlusion scene,and can effectively achieve the depth information acquisition of the array image.The main research contents of this paper include the following points.Firstly,introduce the array camera system,the hardware,and the software in detail.Introduce the array camera which is used to collect and process the array images in detail.Improved array images' stitching method,the stitching speed is improved,and the improved minimum is used.Effectively eliminate the "ghost" phenomenon in the image fusion process,and proves that the array camera can not only be used for depth information acquisition,but also has a broader application prospect.Secondly,introduces the calibration algorithm of multi-camera calibration,selection and calibration accuracy measure of the calibration board,obtained cameras' internal and external parameters.Using the array camera to take pictures of the actual scene,perform stereo corrections on the images acquired by the array camera,using the belief propagation algorithm and the graph cut algorithm to obtain the disparity information of the actual scene,and the advantages and disadvantages of the two algorithms are compared according to the results of the experiment.Then the method of converting disparity information into depth information is introduced.The three cameras pair each other with two adjacent cameras to obtain two disparity images and convert them into depth images,the depth images of the scenes captured by the leftmiddle cameras and the middle-right cameras,extract the contours of these two depth images,calculates the precise horizontal translation distance between the depth images,and use the weighted fusion method to perform image fusion on the two depth images.The experiment shows that the depth map of the fusion is more accurate and effectively solves the occlusion problem compared with the depth map before the fusion.Finally,this paper summarizes the depth information acquisition algorithm based on array camera and proposes the improvement direction that needs to be breakthrough in the future.